Abstract
Multi-agent collaboration holds great promise for enhancing the safety,reliability, and mobility of autonomous driving systems by enabling informationsharing among multiple connected agents. However, existing multi-agentcommunication approaches are hindered by limitations of existing communicationmedia, including high bandwidth demands, agent heterogeneity, and informationloss. To address these challenges, we introduce LangCoop, a new paradigm forcollaborative autonomous driving that leverages natural language as a compactyet expressive medium for inter-agent communication. LangCoop features two keyinnovations: Mixture Model Modular Chain-of-thought (M$^3$CoT) for structuredzero-shot vision-language reasoning and Natural Language Information Packaging(LangPack) for efficiently packaging information into concise, language-basedmessages. Through extensive experiments conducted in the CARLA simulations, wedemonstrate that LangCoop achieves a remarkable 96\% reduction in communicationbandwidth (< 2KB per message) compared to image-based communication, whilemaintaining competitive driving performance in the closed-loop evaluation. Ourproject page and code are at https://xiangbogaobarry.github.io/LangCoop/.